a Vm ‘engine’ that
makes a Difference
By Carl Waldspurger
the PASt DecADe has witnessed a renaissance in server virtualization, which is
transforming enterprise computing by
offering new capabilities and efficien-cies. The following paper by Diwaker
Gupta et al. presents a novel approach
for significantly improving the efficiency of virtualized servers. Their “
Difference Engine” eliminates memory redundancy by exploiting similarity both
within and across virtual machines.
A virtual machine (VM) is a software
abstraction that behaves like hardware.
The classic definition by Popek and
Goldberg is “an efficient, isolated duplicate of a real machine.” For example,
a VM that presents the illusion of being
a physical x86 server may run an unmodified operating system designed
for that platform, such as Windows or
Linux. Neither the OS nor its users need
be aware they are interacting with a VM
instead of dedicated hardware.
Little more than a decade ago, virtual machines were considered a fairly
exotic mainframe technology. Today,
VMs are pervasive in corporate data-centers, and serve as the foundation
for cloud-computing platforms. The
commercial success of virtual machines has influenced the design of
high-volume processor architectures,
which now contain special-purpose
hardware to accelerate virtualization.
Why have VMs proliferated so rapidly? One reason is that virtualization
is an extremely versatile technology.
There is a well-known adage: “All
problems in computer science can
be solved by another level of indirection.” The virtualization software layer,
known as a hypervisor, provides this
level of indirection, decoupling an
OS and its applications from physical
hardware. Eliminating the traditional
“one machine, one OS” constraint
opens up numerous possibilities.
Initially, the most compelling use of
VMs was basic partitioning and server
consolidation. In typical unvirtualized
environments, individual servers were
grossly underutilized. Virtualization al-
lowed many servers to be consolidated
as VMs onto a single physical machine,
resulting in significantly lower capi-
tal and management costs. This abil-
ity to “do more with less” fueled the
rapid adoption of virtualization, even
through economic downturns.
Carl Waldspurger ( firstname.lastname@example.org) is a Principal
engineer at VMware, Palo alto, where he oversees core
resource management and virtualization technologies.